Test program- 3Dmark2001SE
(I was testing Directx functions)

AMD 3000+ Barton, Win98SE, all cards used same 3dfx driver. Motherboard is a rocking Epox with full hardware monitoring. The only cards that would report 4X AGP and 1.5 volt supply were the 4500/4800 AGP boards from 3dfx.

Everything else defaulted and locked to 2X/3.3v

Board | Interface | AGP mode | AGP volts | 2001SE

4500 32 MB,     AGP, 4X, 1.5 V, 4222 points
4800 64 MB,     AGP, 4X, 1.5 V, 4188 points
P-color TV,          AGP, 2X, 3.3 V, 4219 points
P-color L-card,  AGP, 2X, 3.3 V, 4253 points
5500 AGPx4,    AGP, 2X, 3.3 V, 3692 points
5000 AGPx4,    AGP, 2X, 3.3 V, 3059 points

4500 32 MB,    PCI,  2775 points
5500 64 MB,    PCI,  2769 points

For fun since the L-card was the firecracker of the group it was overclocked to 183 mhz, 4320 points was the outcome.

The dual chip boards were run twice to verify results, I'd say 3dfx drivers under SLI mode and Direct X are not so hot. The PCI cards really were pounded.

As you can clearly see the dual chip "4X AGP connector" cards are locked in 2X mode with 3.3 V AGP supply feed, running one of these on a modern board for any period of time would be very bad as they would fit but would be supplied the wrong voltage.



A prototype 3dfx "L" board has been obtained and tested for AGP volts and AGP mode. It is a 3.3V part limited to 1/2x AGP so don't try running it on a modern board if you get one!

Quote:


Well it is a synthetic benchmark and Direct X to boot, we all know under glide/opengl that a 5500 will tear up a 4500 pretty darn quick without breaking a sweat.

3dfx's DirectX drivers were not thought to be the best and there is rumor of 2001SE being biased toward Nvidia hardware..

I thought it was intresting the fastest was a AGP 2X 4500 all else being equal.

yes a test of daytone will be also very interesting.

So that  explains, why is oschars card does not work anymore.

The interesting thing, is that the AGP cards are quite faster than the PCI cards, although the rumors says, that they are both using the same PCI interface and bus clock. (66 mhz ? )

That's indeed odd, I could swear that in all Direct3D Applications, my Voodoo5 5500 64MB AGP always beat any Voodoo4 4500, never saw it the way around.

Unreal 1 or Unreal Tournament (Intro Loop) is pretty good for Glide Benchmarking

Quote:


I was just doing some Direct X testing of VSA-100 using 2001SE and verifing AGP modes and voltages, this wasn't a full blown benchmark session.

I do have a page up on the site with a page of benchmarks that was done months ago.

http://www.thedodgegarage.com/3dfx/bench.htm

Quote:


The test program one is well known to be negative bias in regards to Voodoo hardware, remember the real world vs 2000/2001SE article that was posted on Suidan hardware?

Quote:


Regular PCI slots run at 33 mhz, at slowest setting AGP runs at 66 mhz. The AGP slot has twice the data transfer rate speed.